Institutional change and productivity growth in China's manufacturing: the microeconomics of knowledge accumulation and "creative restructuring"

(1)

Institutional change and productivity growth in

China’s manufacturing: the microeconomics of

knowledge accumulation and “creative

restructuring”

Xiaodan Yu

1,

*, Giovanni Dosi

2

, Jiasu Lei

3

and Alessandro Nuvolari

4 1

Scuola Superiore Sant’Anna, Piazza dei Martiri della Liberta`, 33, 56127 Pisa, Italy. email: x.yu@sssup.it,

2

Scuola Superiore Sant’Anna, Piazza dei Martiri della Liberta`, 33, 56127 Pisa, Italy. e-mail: giovannidosi@

sssup.it,

3

Tsinghua University, Beijing, China. e-mail: leijiasu@163.com and

4

Scuola Superiore Sant’Anna,

Piazza dei Martiri della Liberta`, 33, 56127 Pisa, Italy. e-mail: alessandro.nuvolari@sssup.it

*Main author for correspondence.

Abstract

This article investigates the microeconomics underlying the spectacular growth of productivity in China’s manufacturing sector over the period 1998–2007. Underlying the aggregate evidence of such dramatic growth, one observes a large, albeit shrinking, intra-sectoral heterogeneity coupled with an even more important process of learning and knowledge accumulation. A major process of both catching-up and dying out among the least efficient ones occurs. Furthermore, we explore the effect of the characteristics of firms according to the ownership and governance structure upon the product-ivity distributions, highlighting the importance of the transformation of domestic firms as drivers of technical learning. In essence, China’s fast catching-up process entails more of learning and “creative restructuring” of domestic firms rather than sheer “creative destruction” and even less so a multina-tional corporation-led drive.

JEL classification: O1, O3, O4

1. Introduction

This article analyzes the microeconomics of the spectacular productivity growth in manufacturing in China over the period 1998–2007, disentangling the contribution of the various types of organizational carriers, distinguished according to their institutional characteristics, related to ownership and governance structures (e.g. State-owned, for-eign multinational corporations [MNCs], private-owned). Exploiting a unique firm-level database, we investigate the micro evidence on the dynamic of labor productivity distributions across and within groups underlying the revealed dramatic increase of China’s productivity in manufacturing.

The first major question we address regards the micro statistical properties of the process of industrial develop-ment. How diverse are the degrees of technological efficiency across firms within each industrial sector? And how do they change throughout the development process? In doing that, we study the relative importance as drivers of

doi: 10.1093/icc/dtv011 Advance Access Publication Date: 17 April 2015 Original Article

at Scuola Superiore Sant?Anna on November 29, 2016

http://icc.oxfordjournals.org/

(2)

productivity growth of (i) idiosyncratic, firm-specific learning (the so-called within effect), (ii) competitive selection among incumbents (the so-called between effect), and (iii) entry and exit.

Second, we focus on the institutional “Identity Cards” of the firms involved. Who are the major drivers of the catching-up dynamics? Domestic firms? Foreign direct investments? And among the formers, what is the relative role of State-owned enterprises (SOEs), public–private partnerships, and greenfield private firms?

Concerning the empirical evidence, an extremely robust feature emerging from our study is the high degree of het-erogeneity in production efficiency at all levels of sectoral disaggregation. This is indeed a striking aspect of the devel-opment process but also a persistent one of “mature” industrial dynamics. However, the support of the labor productivity distributions shrinks over time. The degree of productivity dispersion among low-productivity firms diminishes, while that of high-productivity ones increases slightly, reflecting both catching-up by the former and a forging-ahead some of the latter.

Indeed, the dynamics in the support of productivity distributions and especially in the left tail is not just the sheer effect of market selection, but also the outcome of deeply rooted patterns of firm-level learning and institu-tional changes within the corporate sector. The contribution to productivity growth of various types of firms sig-nificantly changes over time. To analyze it, we compare labor productivity distributions and their dynamics across six major ownership types: State-owned, collective-owned, foreign-invested, shareholding, and domestic private enterprises.

In general, foreign-invested enterprises (FIEs) enter with a relatively high productivity, but their contributions change relatively little thereafter. Conversely, all other types of firms display a pronounced change in both the mean and the support of the productivity distribution, showing learning and “catching-up,” particularly during the period of the deepening of institutional reforms. The transformations of State-owned, “shareholding” (public–private joint ventures [JVs]), and private enterprises significantly affect the decreasing dispersion of labor productivity especially among the less-efficient firms, but also contribute to the fattening of upper echelon. Moreover, a (small) group of pri-vate enterprises also contributes to catching-up to the technology-frontier but to a relatively lower degree and mostly in “lower tech” sectors.

The rest of this article is organized as follows. Section 2 reviews some of the incumbent literature on China’s in-dustrial productivity. Section 3 briefly depicts the changing institutional context from 1978 to 2007. Section 4 intro-duces China’s micro manufacturing data set. Next, Section 5 discusses the broad microeconomic picture underlying the striking overall productivity growth. Section 6 analyzes the micro-evidence on labor productivity at sectional level in terms of levels and growth. Finally, Section 7 investigates the properties of such distributions and their dy-namics across different ownership categories. Section 8 concludes.

2. China’s contemporary growth

China’s GDP per capita increased from 524 in 1980 to 5239 dollars in 2007 (PPP constant price and exchange rate, 2005 dollars, cf.The World Bank, 1997). Productivity growth was a major driver of overall economic growth, as also indicated by several recent growth accounting exercises (Bosworth and Collins, 2008;Zheng et al., 2009;Brandt

and Zhu, 2010).

With respect to the aggregate evidence,Young (2003)estimates that labor productivity growth has been 2.6% per year between 1978 and 1998, for the non-agriculture sector. For the manufacturing sector,Szirmai et al. (2005)

andWang and Szirmai (2008)estimate a growth of 2.3% between 1980 and 1990, and 15.9% per year during the

period 1990–2002, witnessing a rapid and accelerating process of catch up.

Brandt et al. (2012)—on the grounds of longitudinal data set rather similar to the one used below—finds an

im-pressive growth rate of total factor productivity (TFP) of 7.7% yearly in manufacturing over the period 1998–2006.1

There are several policy and institutional drivers of the productivity performance of Chinese manufacturing, many of which we can only telegraphically mention here. They include the reform of SOEs; industrial policies affect-ing investment, innovation, finance, and trade; competition policies; and the expansion of private companies. (For a thorough discussion, whose spirit we broadly share, seeDahlman, 2009.) The end-result of the foregoing policies and institutional changes yielded contributions to productivity growth which have been significantly different among

1 Notwithstanding the limitation of such measure of productivity (for a more detailed discussion cf.Dosi and Grazzi, 2006).

(3)

different types of enterprises and across different periods. Such differences are also at the center of the analysis that follows.

In the early stage of reform,Goodhart and Xu (1996)show that township and village enterprises (TVEs) were the main engine of Chinese productivity growth during early 1990s, although starting from very low levels.Jefferson and

Rawski (1994)estimate that the annual growth rates of labor productivity of TVEs were the highest between 1980

and 1992, and conversely in the State sector it was the lowest. More precisely, the growth in the State sector fluctu-ated between 3.8% and 6.3% during the period 1978–1992, peaking in the years 1984–1988, while the productivity of TVEs grew at 5.8% between 1980 and 1984, then increased to 14.4% (1984–1988) and to 17.7% (1988–1992).

In late 1990s,Deng et al. (2005)estimate annual labor productivity growth at 20.4% between 1995 and 2003, ac-cording to a large firm-level databank of industrial large and medium enterprises, with a major contribution stem-ming from the conversion of SOEs to “shareholding” ones (Jefferson and Su, 2006): see below.

Wang and Szirmai (2008)decompose the overall labor productivity growth in terms of sectors and forms of

own-ership. Interestingly, in the 80s changes in the sectoral composition of output, i.e. structural change, bear an import-ant contribution to aggregate industrial productivity growth, while, as we shall see, this is not any longer the case in the subsequent period, when firm restructuring is the main driving force. Together, shifts in the weights of ownership categories account a substantial part of productivity growth during the productivity boom in 1990s, with the expand-ing foreign-invested category (MNCs and JVs) takexpand-ing up a good deal of the shift.

The literature on the micro-evidence on China’s productivity and industrial dynamics (cf.Jefferson et al., 2008) finds a paramount contribution to productivity growth seemingly by exiting and entering firms, even if entering led with restructuring by incumbents. Moreover, the literature highlights the importance of ownership on the average efficiency of Chinese industrial firms, with FIEs exhibiting the highest, and SOEs exhibiting the lowest, efficiency levels.

Along similar lines,Dougherty et al. (2007)in the period 1998–2003 show an overall higher productivity of pri-vate firms. Irrespectively of the ownership, and irrespectively of the criteria of measurements, the performance is truly impressive. So, for example,Fu and Gong (2011)try to estimate TFP and find that the latter appears to grow at an average of 4.8% over the period 2001–2005. However, as informative as means and their dynamics are, they hide profound and long-lasting differences across individual firms.

Indeed, there is an emerging tradition of analysis of the microeconomics behind aggregate means and dynamics, pointing at the stylized fact concerning large and persistent heterogeneity in productivity among firms even within the same narrowly defined industries (see the evidence and reviews in Nelson, 1981; Geroski, 1998, 2002;

Bartelsman and Doms, 2000;Ahn, 2001;Foster et al., 2001,2008;Dosi, 2007;Dosi and Nelson, 2010;Syverson,

2011;Dosi et al., 2012). While this literature mostly focuses on developed countries, micro-heterogeneity applies,

even more so, to emerging economies. Note that such persistent heterogeneities bear far-reaching implications for any analysis of industrial evolution and development more generally. First, and straightforwardly, any dynamic ana-lysis ought to concentrate upon the shape of the distributions of whatever firm characteristics and their dynamics over time. Second, one ought to address the drivers of such dynamics. Third (an often neglected point), ample and persistent heterogeneity in the input/output relation at any time renders analyses in terms of “production function” and TFP, quite dubious or even deeply biased.2_{On that we offer some evidence below supporting our choice using}

labor productivities (even if for the skeptical reader we present some TFP estimates in the Appendix).3

Moreover, firms differ also in their ownership, forms of governance, and origins. This will be the other focus of our study. The longitudinal data below will help indeed in highlighting the movements of sector-specific and organ-ization-specific productivity distributions among firms, that is the very microeconomic action behind the dynamics of the observed means.4

2 The point is well argued inHulten (2001)and sophistications do not alter the basic property that if, micro production functions are not identical up to a scalar the TFP estimates are basically meaningless. This adds to even deeper flaws in the production function construct itself [more inDosi and Grazzi (2006), and earlier inPasinetti (1959),Fisher (1971),

Shaikh (1974)among many nowadays neglected others].

3 For a recent work on China using our same data trying to estimate the allocative distortions stemming from inter-firm heterogeneity in a production function framework, seeHsieh and Klenow, 2009.

4 A large literature on transition economies, whose acknowledgment is here beyond search is interested into impact of firms’ institutional setups upon performances. For an early, similarly micro-castered analysis, cf.Brada et al. (1997).

(4)

However, before looking of the details of the data, it is useful to place the transformation of the Chinese industry against the background of at least equally profound institutional and policy changes.

3. The changing institutional context

Unlike the “big bang” transformation of the economy systems of Eastern Europe and the Soviet Union, China has adopted the “gradualist approach” of economic reform. The reforms which occurred within China’s industrial sector since 1978, occurred at three different levels, namely (i) reforms concerning SOEs, (ii) the political economy of non-State sector, and (iii) the “opening up” policy, and in three temporal stages (Qian, 2000,2002;Zou, 2008).

The first stage (1978–1992) featured the transfer of SOEs’ power of control and governance without touching issues of property rights, the stated aim being “putting plan and market on equal footing” (Qian, 2000). The period up to the early 1990s corresponded to distributions of labour productivities generally displaying an efficiency lead of FIEs.5Interestingly, JVs in quite a few sectors display “best-performers” more efficient than “best-performers” of MNCs (evidence available by the authors upon request).

In the second period (1993–2003) of institutional change, the focus of SOEs reform shifted to corporate govern-ance, and the adjustment in their product portfolios.

The first strategy was to establish a “modern corporate system,” involving the transformation of SOEs into mod-ern enterprises with clear property rights, well-defined responsibility and authority, separation of enterprises from the government, and professional internal management (Qian, 1996).

The second policy step was meant to streamline SOEs output structure, highlighted by the slogan “retain the large while release the small.” The central government defined five groups of sectors kept under the control of the State, i.e. national security sectors, natural monopolies, sectors providing public goods or services, strategic natural resources, and key enterprises in “pillar” and high-tech sectors.6_{Note that, since most SOEs were in net loss until 1997, the transition}

of large and medium SOEs to some “market discipline” involved deep and often socially painful reforms.7

The third step involved the “shareholding system” and the development of a “mixed ownership economy” within large SOEs. This meant transforming SOE in proper shareholder-owned firms possibly involving also private, domes-tic, and foreign investors, yielding an ensemble of “shareholding firms” characterized by mixed ownership. In 1994, thousands of SOEs were selected to conduct such an experiment. Among them, 540 (23%) were transformed into mixed shareholding enterprises (SHEs); 909 (38.8%) turned into companies having the State as the only shareholder; the rest did not implement changes in corporate governance and ownership (Zou, 2008). In 1999, the shareholding system was promoted over all large and medium SOEs, with the State maintaining ultimate control rights over the large/strategic SOEs.

A crucial property of the whole process has been the deep transformation but not the destruction of the techno-logical capabilities embodied in incumbent SOEs (a more articulated discussion of the transformation of SOEs in the 80s inGroves et al., 1994).

Thereafter, in the period 1998–2002 “the State retreats and the private sector moves forward” has been the main strategy, which was designed according to the timetable of China’s WTO accession. SOEs were withdrawn from so-called “competitive” sectors, while have been strengthened in “pillar” industries. Together, a major employment shake-out took place (Cao et al., 1999): the number of workers employed in State-owned manufacturing enterprises fell from 35 million in 1992 to 9.8 million in 2002, with most of the decline in late 1990s (NBS, 2003). The restruc-turing and consolidation of large and medium SOEs were mainly conducted by merger with domestic or foreign firms, conglomeration, and initial public offering on the stock market (Cao et al., 1999). Overall the background ex-amples and inspiring objectives of these policies were the Japanese Keiretsu and the Korean Chaebol. Large State-controlled enterprises were and are mainly in resource-based, capital-intensive, and high-tech industries, such as coal, steel, oil, petrochemicals, aluminum, ship-building, and industrial machinery which all qualify for government “preferential policies” (Brandt et al., 2008).

5 Here and throughout we treat MNCs and JVs within the same broad category.

6 Government designated five “pillar” industries: machinery, electronics, petrochemicals, automobiles, and infrastructure construction (The World Bank, 1997).

7 Through workers layoff, merger and reorganization, separation of social corporate-paid services, debt-to-equity swap, etc.: the goal was achieved roughly by the year 2000.

(5)

At the same time a significant process of privatization of small SOEs and COEs at the county level took place, involving between 50% and 70% of small SOEs, dependent on the provinces. In cities, privatization has occurred in two waves and peaked after 2000 (Garnaut et al., 2006). During the first wave, in mid-1990s, more than half of the firms were privatized in three forms, namely, sale to a private domestic or foreign firm; transformation into a limited liability or joint stock company; and “stock cooperatives” which are essentially employee/manager ownership with some features of cooperatives (Cao et al., 1999). The second wave occurred around 1997, with management buyouts as the most frequent mode.

At the same time, a huge amount of private firms have entered the market, particularly into “competitive sectors,” i.e. sectors outside the “pillar industries.” The output share of domestic private enterprises increased from 3% in 1995 to 22% in 2004 (Wang, 2009).

Prior to WTO membership, during the negotiations in the 1990s, China has gradually liberalized trade and FDI regimes (Branstetter and Lardy, 2008).Huang (2003)highlights three patterns in FDI inflows into China and in the corresponding strategies of MNCs. Early on, a significant portion of FDI inflow into China originated from small and medium enterprises typically from China’s neighboring regions—Hong Kong, Macao, and Taiwan—and operat-ing relatively simple and labor-intensive production and assembly processes. Conversely, investments by large Western and Japanese MNCs constituted only a small portion of total FDI flows. Moreover, the former were the drivers of China’s export-processing regime in the 1990s, yielding an increasing share of export in GDP from around 12% in 1990 to 20% in 2000. Second, joint-ventures typically involved foreign firms and SOEs and were explicitly aimed at the acquisition of advanced technologies. Third, the share of wholly foreign owned enterprise relative to equity JVs increases from mid-1990s, due to the gradual enforcement of the Trade-Related Investment Measures linked with WTO accession.

Interestingly, foreign multinationals and JVs most often aim their production primarily to China’s domestic mar-ket. In 2002, within the 10 industries with the highest share of FDI (accounting for around half of the output), two-thirds of their sales went to the domestic market (Brandt et al., 2008). Moreover, the composition of the origin of FDI has shifted from Hong Kong, Macao, and Taiwan regions to Western countries and Japan. At the same time, MNCs’ investments increasingly favored what Pavitt (1984) calls “scale-intensive sectors” (from automobiles to chemicals), “specialized suppliers” (such as machinery), and ICT.

In the third period (2004–) reform of the governance of SOEs went ahead, involving a higher liquidity of their shares and a greater separation between government and corporate management.

Possibly more interesting for the interpretation of the evidence that we shall present below, during this most re-cent period, in “pillar” industries, and especially the “scale-intensive” ones, SOEs and JVs kept a re-central role, not-withstanding some entering attempts by private domestic firms. And they indeed display a considerable dynamism in terms of technology acquisition and productivity growth. This is a dynamism which is shared across manufacturing by a good deal of private firms (especially in relative low-tech industries). Indeed, as we shall see in detail below the period witnesses the strengthening of the role of domestic firms of all types in the process of catching-up.

4. Data

The raw data set includes all industrial firms with sales above 5 million RMB covering period 1998–2007.8_{This data}

set comes from annual surveys of the “above-scale” industrial enterprises conducted by the National Bureau of Statistics of China.9

Out of that, we extracted manufacturing-only firms,10and we cleaned the data in order to eliminate visible re-cording errors, yielding what we call “China Micro Manufacturing” (CMM).11

8 Year 2004 is a census year and firm’s value added of this year is not available in the data set, and thus, we do not use the 2004 survey in this article.

9 Industry is defined to include mining, manufacturing, and public utilities, according to NBS. Five million RMB is approxi-mately US$ 600,000.

10 Manufacturing firms are those with Chinese Industrial Classification (CIC) code between 13 and 43.

11 We drop firms with missing, zero, or negative value-added; and also firms with a number of employees less than 8, since below that threshold they operate under another legal system (Brandt et al., 2012). NBS has modified its

(6)

Table 1shows the number of observations. Finally, we built a balanced CMM including only the firms that are in the sample for all years (Thus, excluding all firms dropping out at any point and all firms entering after the first year of the sample.) just to give an idea of the intensity of “churning” occurring over the period. The persistent firms are 25,557, that is 18.9% of the total in 1998 and 8.3% in 2007.

Here, we use the definition of “capital stock” originally provided by the data set, that is the sum of all vintages of capital still in use recorded at the nominal cost of the investments.

This work focuses on dynamics of labor productivity, measured as constant-price value added per employee, which we understand as a very plausible proxy of efficiency in different sectors (more on that below).12When appro-priate, in the analysis we shall use relative productivities (normalized with sectoral averages).

piðtÞ ¼ log PiðtÞ < log PiðtÞ > (1)

with Pi¼VANii ; <log PiðtÞ > P

i2sectorjlog PiðtÞ

njðtÞ , PiðtÞ labor productivity, VAiðtÞ real value added, Ni(t) number of employees, of firm i at year t; nj(t) number of firms in sector j at year t.13

We further disaggregate firms by ownership and governance structures. According to the firm’s registration sta-tus, we classify firms into seven ownership categories: SOEs; collective-owned enterprises (COEs), Hong Kong, Macao, and Taiwan-invested enterprises (HMTs); FIEs, including in turn foreign MNCs (FMNC), and JVs with a foreign share above 25%, SHEs; private-owned enterprises (POEs); and other domestic enterprises. As shown in

Table 2, 23 registration categories have been aggregated into 7 larger ones, in line with the approach ofJefferson

et al. (2003). Below, however, due to numerosity, we analyze only the first six categories.

Each firm is classified into a sector according to the four-digit Chinese Industry Classification (CIC) that resem-bles the old US SIC system.14

5. A striking productivity growth: the general picture

Table 3shows the annual growth rate of labor productivity of incumbent firms during different subperiods in the

CMM database, highlighting the dramatic growth of productivity in China’s manufacturing. The overall productivity of incumbents grew at 9.98% during 1998–2007. All sectors display positive productivity growth rates, except pet-roleum refining, which had negative growth during the 1998–2002 period.

Table 1. Number of firms in original and modified data sets, respectively

Year Number of obs (original data set) Number of obs (CMM sample) Deleted obs

1998 148,664 135,510 13,145 (9.70) 1999 146,078 134,004 12,074 (9.01) 2000 147,249 136,830 10,419 (7.61) 2001 155,665 147,612 8053 (5.46) 2002 165,801 157,015 8786 (5.60) 2003 181,013 174,932 6081 (3.48) 2005 250,975 243,906 7069 (2.90) 2006 278,667 271,726 6941 (2.55) 2007 312,304 306,327 5977 (1.95)

Note: Numbers in the brackets denote percentages of deleted observations with respect to original data set. Source: Raw data set and CMM.

industrial classification after 2002. The data set used in this article has adjusted the industrial classification before 2003 into the new standard. Since CIC43 has emerged merely after 2002, we do not consider it here.

12 We undertook the same analysis in terms of TFP.

13 Value-added is deflated by four-digit sectoral output deflators, fromBrandt et al. (2012).

14 In 2003, the classification system was revised to make room for further disaggregation for some sectors, while some other sectors were merged. To make the industry codes comparable across the entire period, we followBrandt et al.

(2012)which has constructed a harmonized classification. Our CMM data set spans over 29 two-digit sectors, 161

three-digit sectors, and 424 four-digit sectors.

(7)

Further, note the remarkable differences in productivity growth across sectors, as such circumstantial evidence of significant inter-sectoral differences in absorptive capacities (Cohen and Levinthal, 1989) of “frontier,” generally for-eign, technologies, and of corresponding differences in the average catching-up rates.

Figure 1offers three snapshots of the non-parametric kernel density distributions of labor productivity, together

with the comparison with the corresponding Italian and French ones, as a vivid illustration of the overall technology gaps with two higher income countries.15

At a first glance the readers might find such a comparison as somewhat far-fetched if one has in mind of a “would production function,” possibly multiplied by some country-specific scalar. After all, Chinese wages have been/are at least an order of magnitude lower than Italian and French ones. As a consequence, one would expect to see the three countries on very different positions on such production functions. But is it really the case? If it were so, one would also expect, first, major differences between China, on the one hand, and Italy and France, on the other, in capital/ output ratios—the appropriate proxy for “capital intensities” when “production functions” differ. And of course one should expect strong correlations between labor productivities and capital/labor ratios within each country and within each sector.16_{Premise to the following: the proxies for “capital” are very noisy on Italian and French data}

and just more so on Chinese ones!

Remarkably, what the evidence suggests is rather at odds with the conventional wisdom. First, capital/output ratios also at sectoral levels do not differ very much between China and the two considered European countries (cf.

Table A1): indeed they tend to be higher in China! Second, the within-country, within-sector micro correlations

be-tween labor productivities (VA/L) and capital/output ratios (K/VA), for whatever proxy for K is used, is robustly

Table 2. Aggregation of the 23 registration categories

Code Ownership category Code Registration status

1 State-owned 110 State-owned enterprises

141 State-owned jointly operated enterprises 151 Wholly State-owned companies

2 Collective-owned 120 Collective-owned enterprises

130 Shareholding cooperatives

142 Collective jointly operated enterprises

3 Hong Kong, Macao, Taiwan-invested 210 Overseas joint ventures

220 Overseas cooperatives

230 Overseas wholly-owned enterprises 240 Overseas shareholding limited companies

4 Foreign-invested 310 Foreign joint ventures

Joint ventures 320 Foreign cooperatives

340 Foreign shareholding limited companies Foreign MNCs 330 Foreign wholly-owned enterprises

5 Shareholding 159 Other limited liability companies

160 Shareholding limited companies

6 Private 171 Private wholly-owned enterprises

172 Private cooperatives enterprises 173 Private limited liability companies 174 Private shareholding companies

7 Other domestic (not included) 143 State-collective jointly operated enterprises

149 Other jointly operated enterprises 190 Other enterprises

Source:Jefferson et al. (2003), Annex I.

15 Referees, compared to an earlier version, have asked us to add PPP estimates, of course, PPPs should not matter when estimating outputs of firms on sectors, but the profession is so inspired by an idea that the whole world is a gen-eral equilibrium that they probably thought that PPPs would change the picture, of course, it does not!

16 For intriguing historical evidence on this point, seeAllen (2012).

(8)

negative in China (cf.Table A2) and is mildly negative in Italy and France (statistics available upon request). Putting it another way, labor and capital productivity are strongly positively correlated. Together, third, even within China, labor productivities and capital/labor ratios—as such a proxy of degrees of production mechanization/automation— are basically orthogonal (seeFigure 2for an illustration of a sector out of most).

Overall, the evidence suggests that very little action comes from “moving along isoquants” in response to relative prices.17_{Rather, “best practice” techniques involve a more efficient use of both labor and capital, and relatedly,}

catching-up fundamentally involves improvements on both dimensions. It is a world of complementarities rather than substitution, wherein technology-gaps and learning efforts are both captured by labor productivity differences, quite independently from relative prices, which TFP proxies might well yield a quite distorted picture of the develop-ment process. Indeed, given the inverse relation between labor productivities and capital/output ratios, labor produc-tivities alone turn out to be a robust proxy for the lower bound of “true” efficiency distributions within countries, but also across countries, with the added advantage of avoiding any explicit or implicit hypotheses on interfactor sub-stitutability and capital measurements.

Table 3. Annual growth rate of labor productivity over 1998–2007, and subperiods 1998–2002 and 2002–2007 among “continuing” firms (i.e. firms keeping in the same two-digit sector over the relevant period)

CIC Sector 1998–2007 1998–2002 2002–2007

13 Food processing of agricultural products 11.25 8.55 13.46

14 Other foodstuff 8.87 5.73 11.22

15 Beverages 10.63 6.20 12.80

16 Tobacco 15.29 11.08 10.65

17 Textile 9.79 10.14 10.54

18 Garments, footwear etc. 7.84 4.84 10.57

19 Leather, fur, feather etc. 7.55 6.52 10.29

20 Processing of timber, manuf. of wood, bamboo etc. 11.87 7.61 14.43

21 Furniture 6.19 4.82 10.40

22 Paper and paper products 10.33 9.48 11.75

23 Printing, reproduction of recording media 8.92 7.54 8.17

24 Articles for culture, education and sports 8.39 7.06 10.03

25 Processing of petroleum, cokeries, nuclear fuel 3.77 1.61 6.36

26 Raw chemical materials and chemical products 11.40 10.44 11.85

27 Pharmaceuticals 8.94 10.64 7.53

28 Chemical fibers 10.31 12.05 8.85

29 Rubber 8.80 7.01 9.39

30 Plastics 5.83 6.43 6.24

31 Non-metallic mineral products 12.76 9.86 14.71

32 Smelting and processing of ferrous metals 13.86 12.68 14.14

33 Smelting and processing of non-ferrous metals 12.45 13.73 12.64

34 Metal products 5.44 7.84 4.32

35 General purpose machinery 15.40 13.76 15.72

36 Special purpose machinery 16.23 13.09 15.13

37 Transport equipment 12.67 11.64 13.05

39 Electrical machinery and equipment 9.51 8.84 9.32

40 Communication equipments, computers etc. 5.64 8.37 3.30

41 Measuring instruments and machinery 9.62 9.46 9.57

42 Artwork and other 10.01 9.32 12.70

Average 9.98 8.73 10.66

Source: Our elaboration on CMM.

17 In fact the empirical elasticities of substitutions implied by the negative micro relation between labor productivities and capital/output ratios (i.e. positive correlations between labor and capital productivities) are positive in sign: “isoquants” do not look like the standard isoquants but are more similar to rays out of the origin.

(9)

Figure 1. Empirical density of labor productivities , whole manufacturing of China, France, and Italy, years 1998, 2002, and 2006. Note: The first row—constant 2000 prices and ex-change rates (IMF source); the second row—PPP adjusted price (World Bank source). Source: our elaboration on CMM, INSEE (on France) and ISTAT-Micro 3 (on Italy).

(10)

Granted all that, let us now focus on the micro picture which the data offer and its dynamics. Start noting the dif-ferent upper bounds of the three country distributions, as such an impressionistic proxy of difdif-ferent inter-country lags and leads (together of course with different sectoral compositions of output). Second, the width of the support of the distribution of China is much larger, revealing much greater technological asymmetries across Chinese firms.

The dynamics of catching-up in China’s manufacturing productivity is indeed associated with (i) a rightward move-ment of the mean of the distributions; (ii) a corresponding rightward movemove-ment of the support; and (iii) as we shall analyze in more detail below, a shrinking of the support itself. Labor productivity distribution is asymmetric and left-skewed. The evolving pattern of the left-tail and that of the right-tail are different as well, as the magnitude of left-tail shift toward higher levels of productivity is very significant, compared with a relatively mild movement of right-tail.

Such dynamics matches what in the old development literature was called a “reduction of the dualistic structure economy” composed by a shrinking traditional/relatively backward part of manufacturing and an expanding “modern” one, which however is only just beginning to push “frontier technologies” further.

An important piece of evidence on intra-sectoral asymmetries in efficiency and their changes over time is the top to bottom ratio of labor productivities.Table 4displays the ratio of the 9th decile over the 2nd decile for each sector from 1998 to 2007. The ratios decrease in most of the sectors, indicating a reduction of productivity dispersion, plausibly due both to learning by laggard firms and selection (exit) of worse performers. The ratios are generally lower in “traditional” ones (CIC 17–24 including textile, garments, leather, furniture, paper manufacturing, etc.) and higher in relatively technology-intensive sectors (e.g. transport equipment, electrical machinery, and communica-tion equipment). The ratios drop more rapidly in the first part of the period under consideracommunica-tion which is also a period of retreat of SOEs from the so-called “competitive sectors.” At the same time, the ratios in quite a few “heavy industries” such as petroleum refining, and non-ferrous metals sectors grows, hinting at some sort of persistent “dual-ism” within such industries (note that growing intra-sectoral asymmetries can and often go hand-in-hand with high average growth rates).

How much of the dynamics in overall productivity distribution is due to inter-sectoral relocation of production?

Table 5displays the time series of value-added shares of each two-digit sector in overall manufacturing. It is

remark-able that relatively little structural change has occurred over the period under investigation. So, for example, the

−2 0 2 4 6 8 −5 0 5 10 ln(K/L) ln(VA/L) CIC 131(1998)

Figure 2. Scatterplot of log(V A/L) versus log(K/L) for Corn milling sector (CIC 131), year 1998. OLS regression: Coefficient¼ 0.038 (Standard error 0.029), R2¼ 0.0006, number of observations ¼ 2838. Source: our elaboration on CMM.

(11)

shares of transport equipment, electrical machinery and equipment, and communication equipment, computers etc. are amongst the highest from the start of the period under consideration, and their total share just increases from 20.7% in 1998 to 22.5% in 2007. A sign indeed that China achieves quite early a “modern” industrial structure. Interestingly, this evidence seems to contradict Kuznets’s view of increasing productivity due to structural change, i.e. movements from low-productivity sectors to high-productivity ones also within manufacturing. On the contrary, our evidence suggests that, unlike what happened in the 1980s (cf.Wang and Szirmai, 2008), the movement of the overall manufacturing means is mainly due to sector-specific dynamics.

Of course, the relative stability of sectoral shares at two-digit sectoral level, does not rule out much more turbulence at finer levels of disaggregation within each two-digit sector: indeed, there is very intensive “micro structural change.” However, the evidence marks a difference with other episodes of industrialization and catching-up, in that China ap-pears to be from the period of our observation already quite mature in terms of broad manufacturing structure. For ex-ample, when South Korea had the same real per capita income that China had in 1998, which was 1973 (Maddison’s historical statistics www.ggdc.net/maddison/oriindex.htm), it had a share of around 22% of textile and clothing over total manufacturing (World Development Indicators database), compared to a 1998 Chinese share of 12%.

In the literature a quite common claim is that export and productivity growth go together (possibly with causality running in both directions). China does indeed display a dramatic rise in the share of export in total manufacturing output and coupled with a dramatic growth in productivity. However the latter is not associated with an increased frequency of exporting firms over the total; on the contrary, after a mild surge, the former’s share in manufacturing

Table 4. Ratio of the average labor productivity of the second highest decile over the second lowest one

CIC Sector 1998 2002 2007

14 Other Foodstuff 19.12 12.20 9.04

15 Beverages 14.89 11.82 9.06

16 Tobacco 17.05 22.95 26.44

17 Textile 8.61 7.01 6.07

21 Furniture 9.29 7.15 6.93

29 Rubber 6.56 7.49 7.42

30 Plastics 8.65 7.18 7.02

32 Smelting and pressing of ferrous metals 9.57 8.58 8.40

33 Smelting and pressing of non-ferrous metals 9.70 8.43 12.72

40 Communication equipment, computers etc. 13.52 11.08 8.36

(12)

value added significantly falls (Figure 3). In fact, the final share in 2007 (48%) is lower than that at the beginning in 1998 (51%).

Figure 4shows the labor productivity distributions of exporters and non-exporters for the years 1998 and 2007,

in some selected sectors (chemical, electrical machinery, and communication equipment) which well illustrate a more general pattern. Note that in 1998 exporters have higher level of productivity and their support of distribution is nar-rower than that of non-exporters. However, a significant catch-up of non-exporters takes place, so that in 2007, ex-porters and non-exex-porters have similar productivity distributions and similar widths of support. (We refrain on purpose from making any causality statement.)

5.1 Learning vs. selection

An interesting assessment of the relative contribution to aggregate productivity growth arising from the reallocation of market shares between incumbents and the entry and exit of firms is provided by the decomposition method intro-duced byBartelsman and Doms (2000).

The aggregate labor productivity of sector I at time t is denoted by equation Y

I;t¼

X

i2I

si;tpi;t (2)

where si,t and pi,t represent the employment share and labor productivity of firm i within two-digit sector I.

The variation of sectoral productivity can be decomposed into five components, of which the sum of

Table 5. Contribution of each two-digit sector to the total value added of manufacturing (percentages)

CIC Sector 1998 2002 2007

14 Other Foodstuff 2.07 1.99 1.99

15 Beverages 3.51 2.69 2.01

16 Tobacco 5.70 5.13 3.11

17 Textile 6.39 5.81 5.23

21 Furniture 0.50 0.53 0.69

29 Rubber 1.33 1.12 1.02

30 Plastics 2.35 2.47 2.28

32 Smelting and pressing of ferrous metals 6.47 6.92 9.52

33 Smelting and pressing of non-ferrous metals 2.08 2.24 4.58

40 Communication equipment, computers etc. 7.39 9.64 8.44

Total 100 100 100

(13)

the first three terms are the contribution of continuing firms, and the last two, of entering an exiting firms, respectively. DPI;t¼ X i2C si;t1Dpi;tþ X i2C

ðpi;t 1 PI;t1ÞDsi;tþ

X

i2C

pi;tDsi;t

þX

i2N

si;tðpi;t PI;t1Þ

þX

i2X

si;t1ðpi;t1 PI;t1Þ:

(3)

The first term represents the within effect, i.e. within firm productivity changes weighted by initial labor shares. This term reflects productivity growth arising from the improvement of the productivity of installed production cap-acity (due to process of learning at firm level). The second term is a between effect—changing labor shares weighted by the deviation of initial firm productivity and initial industry productivity. This term reflects the contribution to productivity growth from the reallocation of market share from less productive to more productive firms (and vice versa). The third term is an interaction effect, capturing the covariations between firm productivities and shares. The fourth term represents the entry effect—a sum of a year-end labor share weighted deviation of productivity of enter-ing firms and initial industry productivity. Finally, the fifth term is an exit effect—a sum of an initial-year labor share weighted difference between initial productivity of exiting firms and initial industry productivity.18

Given the high rates of input and output reallocations between incumbents, and via entry and exits, it is of great importance to investigate the contribution of micro dynamics to aggregate productivity growth.Tables A3andA4 re-port the decomposition results for the whole manufacturing and for each two-digit sector using firm-level data set.

The within effect, i.e. intra-firm productivity growth, accounted for the largest proportion of manufacturing prod-uctivity growth (see the last rows of the tables), that is 47% (1998–2002) and 57% (2002–2007). Net entry, i.e. the overall effect of genuine entry and exit, also explained a substantial amount of growth: 19% (1998–2002) and 34%

Figure 3. Annual percentage contribution of exporters and non-exporters to value added of manufacturing. Source: our elaboration on CMM.

18 Notice that, the between-firm, entry, and exit terms use the deviation between firm productivity and the industry aver-age in the initial period. A continuously operating firm with an increasing share makes a positive contribution to aggre-gate productivity only if it initially has higher productivity than the industry average. Entering (exiting) firms contribute only if they have higher (lower) productivity than the initial average. This treatment of births and deaths ensures that the contribution to the aggregate does not arise because the entering firms are larger than exiting firms, but because of productivity differences.

(14)

0.001 0.01 0.1 1 −8 −6 −4 −2 02468 e CIC 37 (1998) 0.001 0.01 0.1 1 −8 −6 −4 −2 02468 CIC 37 (2003) 0.001 0.01 0.1 1 −8 −6 −4 −2 0 2 4 6 8 exporter non−exporter exporter non−exporter

exporter non−exporter exporter non−exporter exporter non−exporter exporter non−exporter CIC 37 (2007) 0.001 0.01 0.1 1 −8 −6 −4 −2 0 2 4 6 8 CIC 39 (1998) 0.001 0.01 0.1 1 −8 −6 −4 −2 0 2 4 6 8 CIC 39 (2003) 0.001 0.01 0.1 1 −8 −6 −4 −2 0 2 4 6 8 CIC 39 (2007) Figure 4. Empirical density of (log) labor productivity of exporters and non-exporters of transport equipment (CIC 37) and electrical machinery and equipmen t (CIC 39) sectors in selected years (1998, 2003, and 2007). Source: our elaboration on CMM.

(15)

Year 02−07 −80 −60 −40 −20 0 20 40 60 80 100 120 Within Between Interaction Net Entry % share Year 98−02 −80 −60 −40 −20 0 20 40 60 80 100 120 Within Between Interaction Net Entry % share Figure 5. Decomposition by sectors cf. the corresponding Tables A3 and A4 . Note: two-digit sectors are in ascending order (in terms of total productivity growth rate).

(16)

(2002–2007). At a finer level of disaggregation, within and net entry effects are the most important contributing factors to sectoral-level productivity growth (as also shown byFigure 5displaying the relative contribution of for-mula 3). Remarkably, this property holds invariantly across all sectors. Such evidence sheds light on the prominence of both firm’s idiosyncratic learning and “churning” via entry and exit. Conversely, the “between” effect contributed to a smaller fraction of sectoral-productivity growth. Note that the major role of the within effect and the positive contribution of net entry (and especially of its exit component) is a quite robust property of industrial dynamics also in developed countries: for a survey, seeFoster et al. (2001).

6. The dynamics of labor productivity distributions

Let us now focus on the detailed properties of productivity distributions and their dynamics.

The nonparametric kernel densities (Figure 1) provide a general appreciation of the evolutionary patterns of labor productivity distributions. In order to account in more detail for such distributions and their dynamics, we resort to a family of theoretical distributions, Asymmetric Exponential Power (AEP) ones (Bottazzi and Secchi, 2011), which en-ables to properly account for the asymmetries and leptokurtosis of the distributions themselves.19

Note that by estimating an Exponential Power (EP) distribution, one implicitly undertakes robust maximum like-lihood-based comparison of different distribution: indeed for a tail parameter b ¼ 2 on both tails of the distribution approximate a log-normal and for b ¼ 1 a symmetric Laplace (exponential) one (more details inBottazzi and Secchi,

2011).20

6.1 A striking intra-sectoral heterogeneity

Let us study the behavior of normalized (log) labor productivity at a two-digit sectoral level.21_{We begin with the}

analysis of the unbalanced panel, thus investigating the overall properties of the evolving population of Chinese man-ufacturing firms, including entrants and exiters. Later we shall compare our results with the balanced panels of in-cumbents throughout our observation period.

Figure 6 displays the binned probability density, AEP fitted distribution22 _{and, for sake of illustration, the}

Normal fit of normalized (log) labor productivity for selected sectors in years 1998, 2003, and 2007. Over all sectors, AEP estimates display significant deviations from (log-)Normal distributions, and even more so on the low-product-ivity side.

First, the width of the support of both empirical distributions and AEP fit is much larger than that implied by a Normal distribution with a good deal of the deviation due to the wide support of less-efficient firms.

Second, distributions are leptokurtic, with a very fat left tail. Figure 6presents some sectoral examples while

Tables 6and7provide the MLE estimates of the AEP parameters.

Scale parameters aland ar, capturing the left and right width of the support of AEP distributions, are shown in

Table 6. In contrast with the properties of mature economies (see on ItalyDosi et al. (2012)23_{) for the majority of}

sec-tors, al monotonically decreases over time, indicating a shrinking process of the left-tail: this offers another

19 As discussion inBottazzi and Secchi (2011)the AEP distribution is characterized by 5 parameters, two positive shape parameters bland br, describing the tail behavior in the upper and lower tail, respectively, two positive scale

param-eters aland ar, measuring the width of left and right tail, respectively, and one mode parameter m. AEP density is

fAEPðx; pÞ ¼ 1 Ce 1 blj xm aljblhðmxÞþbr1j xm arj br_hðxmÞ (4) where p¼ (bl, br, al, ar, m), h(x) is the Heaviside theta function and where the normalization constant reads

C¼ al A0 (bl)þ ar A0 (br), withAkðxÞ ¼ x kþ1

x1Cðkþ1

x Þ. The AEP reduces to the Exponential Power distribution

(Subbotin, 1923) when al¼ arand bl¼ br. Notice that the smaller the b’s, the fatter the tails.

20 As such, the exercise is of a different kind from estimates of e.g. a Kolmogorov–Smirnov test concerning the diversity of two data-generating processes. Here we try to differentiate families of such processes.

21 For example evidence on distributions, seeBottazzi et al. (2002),Bottazzi and Secchi (2003), andDosi and Grazzi (2006). 22 The five parameters are estimated by maximum likelihood method (MLE) followingBottazzi and Secchi (2011). 23 which finds a quite high stability of aland arover time.

(17)

0.001 0.01 0. 1 1 -1 0 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 CIC26(1998) Empirical bl=1.08,br=1.78 Normal density 0.001 0.01 0. 1 1 -1 0 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 CIC26(2003) Empirical bl=1.04,br=1.93 Normal density 0.001 0.01 0. 1 1 -1 0 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 CIC26(2007) Empirical bl=1.17,br=1.90 Normal density 0.001 0.01 0. 1 1 -1 0 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 CIC40(1998) Empirical bl=0.96,br=1.93 Normal density 0.001 0.01 0. 1 1 -1 0 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 CIC40(2003) Empirical bl=0.88,br=1.87 Normal density 0.001 0.01 0. 1 1 -1 0 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 CIC40(2007) Empirical bl=0.97,br=1.79 Normal density Figure 6. Empirical (log-) density of normalized (log) labor productivity for chemical (CIC26), first row; and communication equipment, computers, etc. (CIC 40), second row; the AEP fit (solid smooth curve) and Normal fit (dotted curve) for year 1998, 2003, and 2007 respectively. Source: our elaboration on CMM.

(18)

circumstantial piece of evidence that labor productivities of less efficient firms tend to regress over time toward the mean. On the contrary, for the majority of sectors, their arincreases over time, indicating that the labor

productiv-ities of more-efficient firms gradually diverge: indicating that some “frontier firms” forge ahead. Overall the support of sectoral labor productivity distributions shrinks over time for most sectors, except a few capital- and scale-intensive industries such as chemical fibers, rubber, nonmetallic mineral products, and nonferrous metals (CIC 25, 28, 29, 31, and 33).

Tail parameters bland brof the AEP distributions are shown inTable 7. The values of blare all around one,

con-firming the Laplace-type fat-tail property of the left-side distributions. Conversely, the values of brare not too far

from 2, hinting at a near log-normality of the upper tail. In fact brin quite a few sector is bigger than 2: the “best

practice” frontiers are quite steep.

Next, we split the whole panel among incumbents (firms which persist throughout our window of observation); “genuine entrants” (firms that are actually born in the year when one starts observing them), “entrant in the panel” (i.e. firms that were born before but only in a particular year enter the panel by overcoming the minimum sale

Table 6. (Unbalanced panel) Estimated al, arparameters and standard errors (in parentheses) for the distribution of labor productivities, two-digit CIC sectors

Unbalanced panel al ar CIC 1998 2002 2007 1998 2002 2007 13 1.24 (0.03) 1.04 (0.02) 0.88 (0.02) 1.39 (0.05) 1.29 (0.04) 1.35 (0.03) 14 1.59 (0.07) 1.11 (0.04) 0.74 (0.03) 1.04 (0.07) 1.28 (0.06) 1.34 (0.05) 15 1.38 (0.06) 1.13 (0.04) 0.75 (0.03) 1.00 (0.07) 1.11 (0.06) 1.35 (0.05) 16 1.39 (0.31) 1.14 (0.24) 0.40 (0.12) 1.31 (0.38) 1.90 (0.37) 3.50 (0.29) 17 0.91 (0.02) 0.75 (0.01) 0.52 (0.01) 1.07 (0.03) 1.08 (0.02) 1.20 (0.01) 18 0.70 (0.02) 0.61 (0.01) 0.49 (0.01) 1.06 (0.03) 0.92 (0.02) 1.03 (0.02) 19 0.79 (0.03) 0.63 (0.02) 0.40 (0.01) 1.13 (0.04) 1.21 (0.03) 1.29 (0.02) 20 1.09 (0.04) 0.74 (0.03) 0.54 (0.02) 1.16 (0.07) 1.11 (0.04) 1.26 (0.03) 21 1.07 (0.07) 0.80 (0.03) 0.45 (0.02) 0.91 (0.08) 1.06 (0.06) 1.43 (0.03) 22 0.98 (0.03) 0.73 (0.02) 0.56 (0.01) 0.82 (0.04) 0.98 (0.03) 1.20 (0.02) 23 1.13 (0.06) 1.06 (0.04) 0.53 (0.02) 1.22 (0.07) 0.97 (0.05) 1.12 (0.03) 24 0.68 (0.03) 0.66 (0.03) 0.54 (0.02) 1.21 (0.05) 1.03 (0.05) 0.96 (0.03) 25 0.84 (0.06) 0.74 (0.05) 0.81 (0.04) 1.26 (0.10) 1.65 (0.09) 1.47 (0.07) 26 1.03 (0.02) 0.87 (0.02) 0.76 (0.01) 1.11 (0.03) 1.23 (0.03) 1.23 (0.02) 27 1.01 (0.05) 0.96 (0.04) 0.80 (0.04) 1.14 (0.06) 1.17 (0.05) 1.31 (0.06) 28 1.11 (0.11) 0.80 (0.06) 0.63 (0.05) 0.93 (0.13) 0.96 (0.08) 1.49 (0.08) 29 0.75 (0.03) 0.79 (0.04) 0.57 (0.03) 1.08 (0.05) 1.10 (0.06) 1.34 (0.05) 30 0.89 (0.03) 0.73 (0.02) 0.48 (0.01) 1.13 (0.04) 1.14 (0.03) 1.38 (0.02) 31 0.79 (0.01) 0.74 (0.01) 0.66 (0.01) 1.22 (0.02) 1.24 (0.02) 1.35 (0.02) 32 0.95 (0.03) 0.92 (0.03) 0.84 (0.03) 1.14 (0.05) 1.07 (0.05) 1.13 (0.04) 33 0.86 (0.03) 0.85 (0.04) 0.62 (0.02) 1.42 (0.06) 1.13 (0.06) 1.94 (0.04) 34 0.85 (0.02) 0.68 (0.01) 0.46 (0.01) 1.15 (0.03) 1.22 (0.02) 1.39 (0.02) 35 0.96 (0.02) 0.75 (0.01) 0.48 (0.01) 1.02 (0.03) 1.01 (0.02) 1.30 (0.01) 36 1.13 (0.04) 1.00 (0.02) 0.54 (0.01) 1.16 (0.05) 1.05 (0.04) 1.37 (0.02) 37 1.08 (0.03) 0.83 (0.02) 0.50 (0.01) 1.12 (0.04) 1.07 (0.03) 1.34 (0.02) 39 0.90 (0.02) 0.73 (0.02) 0.53 (0.01) 1.25 (0.04) 1.27 (0.03) 1.46 (0.02) 40 0.98 (0.03) 0.81 (0.02) 0.54 (0.01) 1.50 (0.05) 1.41 (0.04) 1.41 (0.02) 41 1.03 (0.05) 0.78 (0.03) 0.48 (0.02) 1.25 (0.07) 1.31 (0.05) 1.66 (0.03) 42 0.90 (0.05) 0.78 (0.03) 0.49 (0.02) 1.13 (0.07) 1.00 (0.04) 1.16 (0.03)

(19)

threshold), exiters (firms which die or are acquired), “others” (firms which for whatever reason fail to report in some years).Table 8reports the frequencies of each type.24

Let us begin by comparing the foregoing properties of the distributions with those displayed by the balanced panel alone, comprising only the firms that are in the sample for all years.25Basically the comparison between the two en-sembles of distributions flags out the distinct impact on the evolution of the productivities of firms’ populations in each sector of entry and exit processes.

Tables 9and10show the maximum likelihood estimators of AEP distributions for each two-digit sector.26The

evidence shows a remarkable pairwise similarity between the distributions. In particular, the tails (proxied by the bl Table 7. (Unbalanced Panel) Estimated bl, brparameters and standard errors (in parentheses) for the distribution of labour productivities, two-digit CIC sectors

UNBALANCED PANEL bl br CIC 1998 2002 2007 1998 2002 2007 13 1.18 (0.04) 1.02 (0.03) 1.27 (0.04) 2.47 (0.11) 2.36 (0.09) 2.41 (0.07) 14 1.45 (0.07) 1.09 (0.05) 1.20 (0.06) 1.74 (0.14) 2.19 (0.14) 2.22 (0.10) 15 1.32 (0.07) 1.09 (0.06) 1.09 (0.06) 1.61 (0.13) 1.77 (0.12) 2.24 (0.12) 16 1.56 (0.39) 1.25 (0.33) 0.67 (0.24) 2.30 (0.71) 2.75 (0.68) 6.85 (2.09) 17 1.05 (0.03) 0.99 (0.03) 0.99 (0.02) 1.77 (0.06) 1.88 (0.06) 2.10 (0.04) 18 0.86 (0.03) 0.93 (0.03) 0.96 (0.03) 1.75 (0.07) 1.61 (0.05) 1.68 (0.04) 19 0.90 (0.04) 0.85 (0.04) 0.89 (0.04) 1.80 (0.10) 1.83 (0.08) 1.71 (0.05) 20 1.02 (0.06) 0.94 (0.05) 1.07 (0.05) 2.03 (0.16) 2.00 (0.12) 2.16 (0.07) 21 1.23 (0.11) 0.89 (0.06) 0.79 (0.04) 1.50 (0.16) 1.97 (0.15) 2.28 (0.10) 22 1.15 (0.05) 0.97 (0.04) 0.91 (0.03) 1.44 (0.08) 1.82 (0.08) 2.09 (0.07) 23 1.31 (0.08) 1.19 (0.06) 0.95 (0.04) 1.94 (0.14) 1.68 (0.11) 1.73 (0.07) 24 0.79 (0.05) 1.03 (0.07) 1.08 (0.06) 2.01 (0.14) 1.84 (0.13) 1.58 (0.07) 25 1.09 (0.11) 0.95 (0.09) 1.07 (0.08) 2.21 (0.24) 2.31 (0.19) 2.10 (0.14) 26 1.08 (0.03) 1.04 (0.03) 1.17 (0.03) 1.77 (0.07) 1.98 (0.06) 1.90 (0.04) 27 1.22 (0.08) 1.10 (0.06) 1.36 (0.09) 1.68 (0.11) 1.98 (0.12) 2.39 (0.13) 28 1.19 (0.15) 0.99 (0.10) 1.07 (0.11) 1.74 (0.31) 1.53 (0.18) 2.89 (0.25) 29 0.85 (0.05) 0.92 (0.06) 1.12 (0.08) 2.00 (0.15) 2.01 (0.16) 2.27 (0.12) 30 1.03 (0.04) 0.94 (0.03) 0.94 (0.03) 1.95 (0.09) 1.95 (0.07) 2.16 (0.05) 31 0.96 (0.03) 1.01 (0.03) 1.08 (0.03) 2.05 (0.06) 2.14 (0.06) 2.22 (0.05) 32 1.02 (0.05) 1.03 (0.05) 1.29 (0.06) 1.78 (0.11) 1.84 (0.11) 1.85 (0.09) 33 0.85 (0.05) 1.10 (0.07) 0.97 (0.05) 2.37 (0.16) 1.85 (0.13) 2.78 (0.11) 34 0.94 (0.03) 0.84 (0.03) 0.92 (0.03) 1.83 (0.07) 1.96 (0.06) 2.10 (0.04) 35 1.04 (0.03) 0.92 (0.02) 0.88 (0.02) 1.69 (0.06) 1.68 (0.05) 2.07 (0.03) 36 1.17 (0.05) 1.02 (0.04) 0.86 (0.02) 1.88 (0.10) 1.75 (0.08) 2.11 (0.05) 37 1.09 (0.04) 0.95 (0.03) 0.85 (0.02) 1.66 (0.07) 1.61 (0.06) 1.95 (0.04) 39 1.03 (0.04) 0.91 (0.03) 0.96 (0.03) 2.11 (0.09) 2.23 (0.07) 2.12 (0.04) 40 0.96 (0.05) 0.92 (0.04) 0.97 (0.03) 1.93 (0.10) 1.75 (0.07) 1.79 (0.05) 41 0.93 (0.06) 0.83 (0.05) 0.75 (0.04) 1.87 (0.15) 1.83 (0.11) 2.37 (0.09) 42 1.21 (0.09) 1.14 (0.06) 1.11 (0.06) 1.79 (0.13) 1.59 (0.09) 1.67 (0.06)

24 InBrandt et al. (2012)the definition of entrants is firms which in the year they are first observed indicate a birth year at

most 2 years before. Empirical patterns are quite similar.

25 Note that the data set is able to track firms which persist even if changing governance forms and thus “institutional category”: they continue to remain incumbents.

26 CIC 16 is not shown due to very few number of observations in the balanced panel.

(20)

Table 8. The number of firms and percentages (in parentheses) of incumbents, genuine entrants, entrants (above the threshold), exiters and “others”

1998 1999 2000 2001 2002 2003 2005 2006 2007 Incumbents 25,557 (18.9) 25,557 (19.1) 25,557 (18.7) 25,557 (17.3) 25,557 (16.3) 25,557 (14.6) 25,557 (10.5) 25,557 (9.4) 25,557 (8.3) Genuine entrants NA (NA) 686 (0.5) 729 (0.5) 2024 (1.4) 1484 (0.9) 3422 (2.0) 7331 (3.0) 8435 (3.1) 12,207 (4.0) Entrants in panel NA (NA) 6329 (4.7) 14,486 (10.6) 31,270 (21.2) 46,029 (29.3) 68,024 (38.9) 161,626 (66.3) 205,564 (75.7) 261,544 (85.4) Exiters 102,934 (76.0) 96,317 (71.9) 91,582 (66.9) 84,497 (57.2) 79,807 (50.8) 73,367 (41.9) 43,923 (18.0) 26,188 (9.6) NA (NA) Others 7019 (5.2) 5115 (3.8) 4476 (3.3) 4264 (2.9) 4138 (2.6) 4562 (2.6) 5469 (2.2) 5982 (2.2) 7019 (2.3) Total 135,510 (100.0) 134,004 (100.0) 136,830 (100.0) 147,612 (100.0) 157,015 (100.0) 174,932 (100.0) 243,906 (100.0) 271,726 (100.0) 306,327 (100.0) Source: Our elaboration on CMM.

Table 9. (Balanced panel) Estimated aland arparameters and standard errors (in parentheses) for the distribution of labor productivities, two-digit CIC sectors

Balanced Panel al ar CIC 1998 2002 2007 1998 2002 2007 13 0.89 (0.08) 0.72 (0.05) 0.92 (0.12) 1.38 (0.12) 1.49 (0.09) 1.42 (0.16) 14 0.84 (0.13) 0.68 (0.07) 0.71 (0.08) 1.09 (0.17) 1.36 (0.12) 1.68 (0.15) 15 0.63 (0.07) 0.85 (0.12) 0.93 (0.09) 1.29 (0.12) 1.07 (0.14) 1.07 (0.13) 17 0.47 (0.02) 0.53 (0.03) 0.52 (0.03) 1.30 (0.05) 1.05 (0.05) 1.18 (0.05) 18 0.53 (0.03) 0.63 (0.04) 0.62 (0.05) 1.04 (0.04) 0.77 (0.05) 0.85 (0.05) 19 0.58 (0.05) 0.67 (0.10) 0.32 (0.04) 1.14 (0.07) 1.12 (0.12) 1.37 (0.08) 20 0.65 (0.08) 0.70 (0.11) 0.52 (0.07) 1.29 (0.15) 1.04 (0.15) 1.46 (0.12) 21 0.62 (0.10) 0.51 (0.08) 0.50 (0.08) 1.44 (0.19) 0.96 (0.13) 1.37 (0.11) 22 0.69 (0.06) 0.73 (0.08) 0.69 (0.05) 0.84 (0.07) 0.80 (0.08) 1.08 (0.08) 23 0.60 (0.06) 0.53 (0.04) 0.59 (0.05) 1.10 (0.09) 1.12 (0.07) 1.00 (0.08) 24 0.62 (0.06) 0.60 (0.10) 0.46 (0.06) 1.06 (0.09) 0.91 (0.13) 0.91 (0.10) 25 0.76 (0.13) 0.51 (0.17) 1.03 (0.21) 1.04 (0.18) 1.35 (0.23) 1.20 (0.23) 26 0.66 (0.03) 0.65 (0.04) 0.67 (0.03) 1.14 (0.05) 1.27 (0.06) 1.35 (0.06) 27 0.68 (0.07) 0.60 (0.06) 0.66 (0.07) 1.31 (0.12) 1.40 (0.11) 1.50 (0.13) 28 1.04 (0.21) 0.46 (0.12) 0.79 (0.61) 0.69 (0.14) 1.57 (0.17) 1.11 (0.46) 29 0.51 (0.09) 0.59 (0.18) 0.65 (0.11) 1.07 (0.14) 1.26 (0.26) 1.27 (0.17) 30 0.62 (0.04) 0.57 (0.04) 0.55 (0.05) 1.24 (0.07) 1.24 (0.08) 1.25 (0.08) 31 0.54 (0.02) 0.59 (0.04) 0.69 (0.03) 1.14 (0.04) 1.15 (0.06) 1.25 (0.06) 32 0.85 (0.08) 0.77 (0.08) 1.25 (0.23) 0.88 (0.10) 0.99 (0.11) 0.82 (0.19) 33 0.57 (0.07) 0.87 (0.12) 0.64 (0.08) 1.64 (0.11) 0.96 (0.16) 1.63 (0.15) 34 0.51 (0.03) 0.56 (0.04) 0.46 (0.03) 1.37 (0.06) 1.32 (0.07) 1.45 (0.07) 35 0.70 (0.04) 0.47 (0.02) 0.57 (0.03) 0.90 (0.05) 1.19 (0.04) 1.33 (0.05) 36 0.75 (0.05) 0.73 (0.06) 0.66 (0.04) 1.19 (0.08) 1.08 (0.08) 1.41 (0.07) 37 0.66 (0.04) 0.73 (0.05) 0.81 (0.06) 1.18 (0.06) 0.95 (0.06) 0.94 (0.07) 39 0.77 (0.06) 0.63 (0.05) 0.56 (0.04) 1.05 (0.08) 1.16 (0.07) 1.64 (0.07) 40 0.69 (0.05) 0.64 (0.06) 0.45 (0.03) 1.54 (0.08) 1.36 (0.09) 1.59 (0.07) 41 0.58 (0.06) 0.65 (0.10) 0.57 (0.05) 1.37 (0.12) 1.30 (0.15) 1.49 (0.10) 42 0.72 (0.07) 0.68 (0.08) 0.62 (0.09) 1.17 (0.12) 0.93 (0.09) 1.03 (0.11)

(21)

and brparameters) are similarly fat-tailed. The only major significant difference concern the alparameters which in

the unbalanced panel start large at the beginning of the period and shrink thereafter, while in the balanced one they tend to be relatively small from the beginning. That is, the shrinking of the width of the left part of the distribution— the less efficient one—is fundamentally due to some disappearance of the relatively less productive among the initial incumbents (and to some extent “churning,” the entry and exit of new short-lived firms).

Figure 7displays the average labor productivity by groups over the years. Note that the productivity of the

incum-bents is always the highest. Interestingly, after 2002, productivity is higher among exiters than entrants (this applies also at two-digit level) hinting at the marginality of a good deal of de novo entrants.Figure 8shows the labor prod-uctivity distribution by type in two selected sectors (all two-digit sectors display qualitatively the same pattern). Not surprisingly the left tails of the distribution for exiters have large supports (many exiters are relatively low-efficiency firms). What is more puzzling is that the right tail is very close to that of incumbents. One of the reasons might be that “exit” may mean acquisition and not death. However, it may well be the case that firms actually die for reasons other than lack of technological efficiency, but e.g. due to financial fragility in presence of a myopic financial system (on the Italian evidence, seeBottazzi et al., 2006). We leave these important questions to further research.

Table 10. (Balanced panel) Estimated bland brparameters and standard errors (in parentheses) for the distribution of labor productivities, two-digit CIC sectors

Balanced panel bl br CIC 1998 2002 2007 1998 2002 2007 13 1.18 (0.14) 0.97 (0.10) 1.42 (0.21) 2.30 (0.26) 2.69 (0.26) 2.34 (0.30) 14 1.42 (0.26) 1.05 (0.16) 1.06 (0.18) 2.07 (0.37) 2.10 (0.28) 2.66 (0.37) 15 1.01 (0.17) 1.33 (0.24) 1.08 (0.16) 1.95 (0.28) 1.53 (0.26) 1.57 (0.25) 17 0.85 (0.07) 0.95 (0.07) 0.97 (0.08) 2.21 (0.14) 2.11 (0.15) 2.08 (0.14) 18 0.76 (0.05) 1.22 (0.12) 1.26 (0.13) 1.86 (0.14) 1.33 (0.12) 1.24 (0.11) 19 0.75 (0.08) 1.38 (0.26) 0.93 (0.16) 1.72 (0.18) 1.62 (0.22) 1.97 (0.21) 20 0.92 (0.18) 1.14 (0.24) 0.73 (0.13) 2.61 (0.51) 1.60 (0.31) 1.99 (0.31) 21 0.93 (0.23) 0.99 (0.23) 0.63 (0.11) 3.21 (0.77) 1.66 (0.34) 2.50 (0.45) 22 1.25 (0.16) 1.40 (0.19) 0.94 (0.10) 1.32 (0.15) 1.37 (0.17) 1.72 (0.18) 23 1.01 (0.14) 0.80 (0.09) 1.02 (0.14) 1.72 (0.21) 2.05 (0.23) 1.63 (0.20) 24 0.84 (0.11) 1.31 (0.27) 1.10 (0.21) 1.89 (0.27) 1.78 (0.31) 1.74 (0.26) 25 1.07 (0.26) 1.34 (0.54) 1.27 (0.34) 1.53 (0.36) 1.82 (0.43) 1.34 (0.34) 26 0.98 (0.07) 1.08 (0.09) 0.98 (0.07) 1.70 (0.11) 2.08 (0.15) 2.03 (0.13) 27 1.15 (0.16) 1.10 (0.16) 1.12 (0.17) 2.24 (0.27) 2.91 (0.35) 3.02 (0.39) 28 1.62 (0.47) 0.57 (0.14) 3.21 (2.51) 1.01 (0.30) 2.44 (0.60) 1.62 (0.71) 29 1.17 (0.26) 1.53 (0.53) 1.22 (0.27) 2.27 (0.41) 2.91 (0.71) 2.01 (0.35) 30 0.93 (0.09) 1.05 (0.12) 1.08 (0.13) 2.11 (0.19) 2.30 (0.22) 1.91 (0.17) 31 0.87 (0.05) 1.16 (0.10) 1.01 (0.07) 1.96 (0.12) 1.87 (0.13) 2.06 (0.14) 32 1.07 (0.15) 1.03 (0.15) 2.19 (0.47) 1.36 (0.21) 1.58 (0.25) 1.57 (0.41) 33 0.67 (0.10) 1.18 (0.22) 0.98 (0.18) 3.34 (0.53) 1.76 (0.37) 2.38 (0.36) 34 0.84 (0.08) 1.05 (0.11) 0.94 (0.11) 2.02 (0.15) 2.03 (0.17) 2.14 (0.17) 35 1.10 (0.08) 0.89 (0.06) 0.91 (0.06) 1.48 (0.10) 2.03 (0.12) 2.27 (0.14) 36 0.99 (0.09) 1.20 (0.13) 0.77 (0.06) 2.09 (0.20) 1.67 (0.17) 1.98 (0.16) 37 0.92 (0.07) 1.14 (0.10) 1.38 (0.14) 1.68 (0.13) 1.44 (0.12) 1.41 (0.13) 39 1.37 (0.14) 1.26 (0.13) 1.02 (0.10) 1.80 (0.17) 1.97 (0.16) 2.50 (0.18) 40 0.87 (0.09) 1.09 (0.13) 0.73 (0.07) 2.07 (0.19) 1.84 (0.18) 2.14 (0.17) 41 0.89 (0.14) 1.22 (0.25) 0.74 (0.10) 2.15 (0.30) 1.84 (0.29) 1.77 (0.20) 42 1.03 (0.15) 1.28 (0.21) 1.22 (0.22) 1.85 (0.26) 1.19 (0.15) 1.50 (0.22)

(22)

6.2 Growth rates of labor productivity

Let us look next at the properties of growth rates.Figure 9displays the empirical distribution and EP fit of growth rates of labor productivity of selected sectors for periods 1998–2002 and 2003–2007.

They are all fat-tailed and symmetric. Such long-term “lumpiness” of productivity growth events clearly militates against any notion of productivity growth resulting from of a smooth process made by small independent improve-ments. Rather, it appears characterized by relatively frequent “big” idiosyncratic shocks (Dosi et al., 2012;Dosi, 2007), plausibly of technological but also strategic and institutional nature.

Table 11shows the autocorrelation coefficients for relative labor productivity in levels and first differences: the

rela-tive productivity levels are quite persistent over time, with only some relarela-tively mild regression-to-the-mean tendency.27

7. The comparative evidence across ownership and governance forms

If we treat firms switching ownership categories as continuing firms (as shown inTable 12) apply formula 3 (cf. above, Section 5), we can decompose the total growth of manufacturing sector into the contribution of each owner-ship and further decompose the latter into the contribution of each effect. Using this method, the contributions of ef-fects can be compared across different ownership types.28

Almost 50% of within effect can be attributed to SOEs only during 1998–2002 and both SOEs and SHEs between 2002 and 2007. The contribution of within effect of the ownership-switching group has been increased from 8.6% to 15.5%. The contribution of genuine entry effects doubled to 26% in the second subperiod. FIEs, SHEs and China’s POEs are the most contributed groups.

Furthermore, at finer level of disaggregation, we replicate the same exercise for each two-digit sector (see

Figures 10and11) and find similar evidence.

7.1 Aggregate evidence

Distributions and their dynamics are deeply affected by the organizational types of the micro entities. Let us consider them.

Figure 7. Average value added per employee of incumbents, entrants, and exiters (at 1998 constant price; 1000 yuan¼ 1). Source: our elaboration on CMM.

27 For the estimates we adopt the approach described inChesher (1979).

28 For a highly complementary analysis on the micro patterns of firm growth by ownership types, seeDuschl and Peng

(2015), this issue.

(23)

0.001 0.01 0.1 1 −4 −2 0 2 4 6 8 10

Incumbents Exiters Entrants

CIC 37 (1999) 0.001 0.01 0.1 1 −4 −2 0 2 4 6 8 10

CIC 37 (2002) 0.001 0.01 0.1 1 −4 −2 0 2 4 6 8 10

CIC 37 (2006) 0.001 0.01 0.1 1 −4 −2 0 2 4 6 8 10

CIC 40 (1999) 0.001 0.01 0.1 1 −4 −2 0 2 4 6 8 10

CIC 40 (2002) 0.001 0.01 0.1 1 −4 −2 0 2 4 6 8 10

CIC 40 (2006) Figure 8. Empirical (log-) density of (log) labor productivity for transport equipment (CIC 37) and communication equipment, computers, etc. (CIC 40). Sourc e: our elaboration on CMM.